Conditional survival and recurrence of remnant gastric cancer after surgical resection: A multi-institutional study

2.1 Population and covariates

Remnant gastric cancer was defined as all new diagnoses of gastric carcinomas in the remnant stomach after partial gastrectomy regardless of the previous disease. From January 2004 to January 2017, 379 patients with RGC were retrospectively identified from seven centers in China (Fujian Medical University Union Hospital, the First Affiliated Hospital of Fujian Medical University, Zhangzhou Affiliated Hospital of Fujian Medical University, Longyan First Affiliated Hospital of Fujian Medical University, the First Affiliated Hospital of Xiamen University, the Affiliated Hospital of Putian University and Shanxi Provincial Cancer Hospital). The inclusion criteria for patients were defined as follows: the presence of remnant gastric carcinoma, no combined malignancy, no distant metastasis, no neoadjuvant chemotherapy and/or radiotherapy, complete basic information and complete survival data. The remaining 298 patients who underwent curative intent resection were included in the present study. The institutional review boards of all participating institutions approved the study (approval number: 2019KY021).

Well-known clinicopathological data were collected routinely. T stages and N stages were classified according to the criteria described in the AJCC staging manual (8th edition).12 The surgical procedure begins with separation of adhesion. D2 lymphadenectomy is performed, followed by resection of the remnant stomach. The reconstruction approach is performed using the Roux-en-Y method. Fluorouracil-based adjuvant chemotherapy was recommended for patients with advanced RGC in all participating institutions. All patients received standard postoperative follow-ups, including visits every 3-6 months for the first 2 years, every 6-12 months from the 3rd to 5th year, and annually thereafter. Most routine follow-up appointments included a physical examination, laboratory testing, chest radiography, and abdominopelvic ultrasonography or computed tomography. All patients were observed until death or the final follow-up date in January 2019. Overall survival (OS) was defined as the time from surgery to death from any cause. Disease-specific survival (DSS) was defined as the time from surgery to death from gastric cancer. In this study, disease-specific death events included death from recurrence and distant metastasis at a specific site, and death after tumor marker elevation but no specific recurrence site was found. In addition, we classified death within 6 months as disease-specific death.

2.2 Conditional survival concept

Conditional survival, the origin of which is rooted in conditional probability from biostatistics, can be calculated using the life-table method.16 The 3-year conditional survival (CS3) at x years indicates the likelihood of an additional 3-year survivorship for an individual who has already survived for x years after the initial treatment and is calculated as follows: CS3 = S(x + 3)/S(x).20 In this study, the data for overall survival and disease-specific survival were used to calculate the 3-year conditional overall survival (COS3) and 3-year conditional disease-specific survival (CDSS3), respectively.

2.3 Definition and categorization of recurrence

Recurrence was defined as the presence of a biopsy-proven tumor showing adenocarcinoma cells or the presence of imaging features highly suspicious of tumor recurrence. Recurrences were categorized by the site involved, as follows: locoregional, peritoneal, distant or multiple. The presence of recurrent disease in two or more sites was defined as multiple. Multiple recurrences in the same site were not categorized as “multiple” sites of recurrence.

2.4 Statistical analysis

The associations between variables and OS or DSS were assessed using a Cox proportional hazards model. CS analysis was then used to assess the possible changes in the prognostic impact of the clinicopathologic variables over time after surgery. The differences in CS between groups were compared via the standardized differences (d) method, which was first described by Cucchetti et al21 and has been employed by several groups,14, 22 where |d| < 0.1 indicates very small differences between groups, 0.1 ≤ |d| < 0.3 indicates small differences, 0.3 ≤ |d| < 0.5 indicates moderate differences and |d| ≥ 0.5 indicates obvious differences. The associations of relevant clinicopathological variables with recurrence were examined using logistic regression. In addition, we applied a second multivariable analysis at later time points (time-dependent multivariable analysis) to assess the independent predictors of survival or recurrence among patients who were alive after a certain number of years.23

All data were processed using SPSS 22.0 (SPSS) and R software (version 3.5.2). All tests were two-sided with a significance level set to P < 0.05.

3.1 Demographic and clinicopathological characteristics

Table S1 lists the general clinicopathological data of 298 patients with RCG who underwent radical surgery; 80.5% of patients were diagnosed with cancer in the remnant stomach more than 5 years after the first gastrectomy. The mean age of diagnosis for RCG patients was 64.9 years (range 37-89), and the ratio of male to female patients was 7.3:1. In total, 158 (53.0%) RGC were in the anastomotic area, and 140 (47.0%) RGC were in the non–anastomotic area. A total of 51.7% of the 298 patients were diagnosed with pathological stage III disease. The average number of lymph nodes (LN) retrieved was 16.0 (range 1-59), and the number of LN retrieved in 54.0% of patients was no more than 15.

3.2 Actual survival

After a median follow-up time of 61 months, 107 patients (35.9%) had recurrence, 157 patients (52.7%) died, of which 138 patients (87.9%) experienced disease-specific death. The 5-year OS of the whole group was 41.2%, and the 5-year DSS was 45.8% (Figure S1A,B). The risk of death after RGC radical surgery was not constant, with most RGC patients experiencing all-cause death or disease-specific death in the first 3 years after surgery (Figure S1C,D). When assessed over time, the likelihood of all-cause death and disease-specific death peaked at 12 months after surgery and diminished thereafter.

We also evaluated the probability of RGC patients surviving for a certain period after surgery (Figure 1A) and showed that with the prolongation of postoperative survival, the probability of disease-specific survival improved. Correspondingly, with the prolongation of postoperative survival time, the probability of recurrence in RGC patients decreased (Figure 1B).

3.3 Comparing actual survival with conditional survival

Figure 2A shows the actual OS of the whole group of patients and the annual COS3 changes within 5 years after surgery. The results showed that for the entire cohort of RGC patients, the 3-year actual OS and the COS3 at baseline were equal; however, contrary to the declining trend of the actual OS curve, the COS3 increased with prolonged postoperative survival. After 2 years of survival, the COS3 was 59.3%, which was different from the actual 5-year OS (41.2%). When the patient survived for 5 years, the COS3 reached 84.0%, which was significantly higher than the actual 8-year OS (34.6%). A similar trend can be observed for DSS (Figure 2B). For example, when a patient survives for 5 years after surgery, the CDSS3 is 89.8%, which is significantly higher than the actual 8-year DSS of 41.2%.

3.4 Univariate and multivariate cox regression analysis at baseline

Univariate Cox regression analysis (Table S2) showed that for the 298 patients with RGC, age, sex, lymphovascular invasion, combined resection, histological differentiation, macroscopic type, tumor size, T stage, and N stage were associated with OS at the time of surgery (P for all < 0.05). Previous disease, the interval between the first operation and the RGC diagnosis, and whether the number of LN retrieved was greater than 15 were not associated with OS at baseline (P for all > 0.05). Further multivariate Cox regression analysis showed that age (≥65 years: hazard ratio [HR] = 1.41, P = 0.038), poorly differentiated adenocarcinoma (undifferentiated: HR = 1.50, P = 0.018), large tumor size (≥4 cm: HR = 2.28, P < 0.001), advanced T stage (≥T3: HR = 3.40, P < 0.001) and lymph node metastasis (N+: HR = 2.42, P < 0.001) were independent risk factors for OS at baseline. After adjusting for the potential confounding factors for DSS, multivariate analysis showed that histological differentiation, tumor size, T stage and N stage were independent prognostic factors that initially affected DSS.

3.5 Stratified analysis of 3-year conditional survival changes

Figure 3 shows the temporal trends of OS and COS3 for each independent prognostic factor in RGC patients. The stratified analysis found that for any subgroup, COS3 was higher than the actual OS, and this gap showed an increasing trend over time. It is worth noting that in RGC patients who initially had unfavorable prognostic features, the gap between the actual OS and COS3 was more pronounced over time. For example, for patients with ≥T3 stage disease, although their 8-year OS was only 25.4%, their COS3 reached 83.0% (Δ57.6%) after 5 years of survival. However, the trend corresponding for patients with the initial relatively favorable prognostic features was not obvious, such as ≤T2 patients who had an 8-year OS of 60.1% and a COS3 of 85.7% (Δ25.6%) 5 years after surgery. Subgroup analyses (Table S3) revealed that the initial unfavorable prognosis factors (large tumors, undifferentiated cancer, advanced T stage and advanced N stage) corresponded to a rapid increase in COS3. We also found that the difference in COS3 between the patients with initial unfavorable prognostic features and those with favorable prognostic features gradually decreased over time. For example, the d values between ≥T3 patients and ≤T2 patients gradually decreased from 0.72 (significant difference) at baseline to 0.37 (moderate difference) at 3 years postoperatively and 0.07 (very small difference) at 5 years after surgery.

Similarly, patients with initially unfavorable tumor characteristics had the most significant changes in CDSS3 compared to patients with initially favorable tumor characteristics; differences in long-term prognosis between the calculated CDSS3 and actuarial DSS were again most pronounced among patients who were initially predicted to have the worst DSS (Figure S2, Table S4).

3.6 Timing and pattern of recurrence

Most recurrences (more than 90.0%) occurred within 3 years after surgery (Table S5). Figure 4A shows the distribution of initial recurrence sites in all patients. Most patients (74.8%) had initial recurrence involving only one site; the remaining 27 (25.2%) patients had multiple sites of recurrence. Figure 4B-D depicts the recurrence pattern in each staged RGC patient; the pattern showed that patients with RGC are more prone to recurrence at multiple sites as the disease stage progresses (Phase I: 10.0%, Phase II: 14.3%, Phase III: 30.2%; P < 0.001). There was a significant difference in recurrence rates between stages (P < 0.001), with a recurrence rate of 15.4% in stage I, 26.6% in stage II, and 49.4% in stage III. The logistics regression showed that only the AJCC stage was an independent factor that affected postoperative recurrence in RGC patients (Table S6). Detailed analysis of the AJCC staging showed that advanced T stage (T4: OR = 3.58) and advanced N stage (N3: OR = 2.99) were independent risk factors for postoperative recurrence in RGC patients (Table S7). Among patients who experienced a recurrence, the median post–recurrence survival was only 8.0 months. There was no difference in survival after relapse in each stage (P = .700; Figure S3).

3.7 Dynamic changes in risk factors for survival and recurrence

Time-dependent multivariate Cox analysis (Table 1) showed that at baseline, age, histological differentiation, tumor size, T stage and N stage were independent prognostic factors for OS. After 3 years of survival, only age and tumor size were independent prognostic factors for OS (P for all < 0.05). Notably, after 4 years of survival, age and tumor size no longer independently affected OS in RGC patients (P for all > 0.05). At baseline, histological differentiation, tumor size, T stage and N stage significantly affected DSS, but after 3 years of survival, only tumor size remained a significant predictor of DSS (P < 0.05). However, after 4 years of survival, the common prognostic factors could no longer independently affect DSS (P for all > 0.05).

If the AJCC stage was included in the time-dependent multivariate regression analysis, then only the AJCC stage and age were independent prognostic factors that affected OS at baseline (Table S8). The AJCC stage can continuously affect OS in the first 3 years after surgery (P < 0.05). However, after 4 years of survival, the AJCC stage no longer independently affected the OS of RGC patients. For DSS, after a stepwise backward variable removal, only the AJCC stage independently affected the OS postoperatively up to 3 years after surgery (P < 0.05), but after 4 years of survival, the AJCC stage lost its prognostic significance (P > 0.05).

Further time-dependent logistic regression analysis (Table S9) showed that the AJCC stage independently affected recurrence in RGC patients within 2 years after surgery (P < 0.05). In contrast, for patients who survived disease-free for more than 2 years, there was no significant difference in the probability of recurrence among the three stages (P for all >0.05).

3.8 Follow-up models

Based on the analyses of conditional survival and recurrence, we developed a postoperative follow-up model for RGC patients of different stages (Figure 5). For patients with stage I disease, follow-up visits are recommended every 6 months for the first 3 years after surgery. In the 4th and 5th years after surgery, follow-up visits are recommended once annually. For patients with stage II-III disease, follow-up visits are recommended every 3 months in the first 2 years after surgery, every 6 months in the 3rd year after surgery, and annually in the 4th and 5th years after surgery. For RGC patients who have survived for more than 5 years after surgery, the visits should be based on routine healthcare programs and supplemented by specialist examinations related to GC. At the same time, during the follow-up period, clinicians can adjust the corresponding follow-up strategy according to the patient’s symptoms and imaging findings.